Claude Academy

Claude Code as a Team Bot

Channels extend Claude Code beyond the terminal. Instead of only interacting through the CLI, Claude can receive and send messages through messaging platforms — Discord, Telegram, and potentially others.

This turns Claude Code into an always-on team AI that responds to messages, answers questions, and performs actions — all triggered by team chat.

How Channels Work

┌──────────┐ ┌─────────────┐ ┌──────────────┐ │ Discord │ ←──→ │ Claude Code │ ←──→ │ Your Code │ │ Telegram │ │ (Session) │ │ (Local) │

└──────────┘ └─────────────┘ └──────────────┘

1. A team member sends a message in Discord/Telegram

2. The message arrives at Claude Code through the channel

3. Claude processes the message with full tool access (files, bash, MCP)

4. Claude sends a response back to the platform

Claude has the same capabilities as in a normal session — it can read files, run commands, query databases — but the trigger is a chat message instead of terminal input.

Setting Up Telegram

Step 1: Create a Bot

1. Open Telegram and find @BotFather

2. Send /newbot and follow the prompts

3. Copy the bot token

Step 2: Connect to Claude Code

claude --channels telegram

Follow the setup instructions to link your bot token. Claude Code connects to Telegram and starts listening for messages.

Step 3: Interact

In your Telegram chat with the bot:

You: What's the current status of our test suite? Bot: All 147 tests passing. Last run: 3 minutes ago. No failures in the last 24 hours. You: Review the latest PR Bot: PR #45 "Add rate limiting to API endpoints" 3 findings: - Medium: Missing rate limit on /health endpoint - Low: Magic number 100 should be a constant

- Suggestion: Add rate limit headers to responses

Use Cases

Team Q&A Bot

Set up a Claude Code bot in your team's Discord/Telegram that can answer questions about your codebase:

Developer: How does the auth middleware work? Bot: The auth middleware (src/middleware/auth.ts) extracts the JWT from the Authorization header, verifies it using the secret from env.JWT_SECRET, and attaches the decoded user to request.user. It returns 401 for missing/invalid tokens. Developer: Where's the payment webhook handler? Bot: src/routes/webhooks/stripe.ts, line 23. It verifies the

HMAC signature, then dispatches to handlers based on event type.

Deployment Monitor

Bot: [Automated] Deployment to staging complete. Health check: PASS Response time: 145ms (normal) Error rate: 0.01% (normal) Developer: Looks good, deploy to production Bot: Deploying to production... [2 minutes later] Production deployment complete. Health check: PASS

All systems nominal.

Code Review Trigger

Developer: /review 42 Bot: Reviewing PR #42 "Implement order notifications"... Security: No issues found Quality: 2 suggestions (variable naming, extract constant) Performance: 1 warning (N+1 query in notification batch)

Full review posted as PR comment.

On-Call Assistant

[3 AM Alert from monitoring] On-call Dev: What's causing the spike in 500 errors? Bot: Analyzing error logs... Root cause: Redis connection pool exhausted. The pool limit is 10 but we have 15 concurrent webhook processors. This started after the 22:00 deployment. Quick fix: Increase pool limit to 25 in config/redis.ts Proper fix: Add connection pooling with backoff On-call Dev: Apply the quick fix Bot: Applied: increased pool limit to 25 in config/redis.ts Committed: "fix: increase Redis pool limit (hotfix)"

Pushed to main. Deploying...

Configuration

Channels configuration is managed through the Claude Code session:

# Start with channels claude --channels # Or add to a running session

/channels

Access Control

Control who can interact with your bot:

Allowlist: Only specific users/groups can interact
DM policy: Whether the bot responds in DMs
Group policy: Whether the bot responds in group chats

Research Preview Considerations

Channels is a research preview feature:

API may change: Configuration and behavior might evolve
Stability: Less stable than core features
Features: Some features may be added or removed

Use it for:

Team productivity experiments
Non-critical monitoring
Development workflow automation

Don't rely on it for:

Production incident response (have a manual fallback)
Customer-facing bots (use a dedicated bot framework)
Security-sensitive operations without additional safeguards

Security Notes

When running Claude Code as a team-accessible bot:

1. Lock down permissions: The bot should have minimal write access

2. Read-only by default: Start with read-only capabilities, add write access carefully

3. Audit logging: Log all bot interactions

4. Scope access: Restrict which channels/users can interact

{
  "permissions": {
    "allow": [
      "Read",
      "Bash(git log:*)",
      "Bash(git diff:*)",
      "Bash(pnpm test)"
    ],
    "deny": [
      "Write(*)",
      "Bash(rm *)",
      "Bash(git push *)"
    ]
  }
}

A read-only bot is safe. A write-enabled bot needs careful permission scoping.

Key Takeaway

Channels connect Claude Code to messaging platforms like Discord and Telegram, creating an always-on AI team member. Messages flow bidirectionally — team members ask questions or trigger actions, Claude responds with full code access. Use cases include team Q&A bots, deployment monitors, code review triggers, and on-call assistants. Channels is a research preview — use it for team productivity and non-critical workflows. Always lock down permissions when exposing Claude through a messaging channel.

Channels: Discord and Telegram Integration

Learning Objectives